FPKM interpretation

[Lalla]

Hi all!

I am analyzing differential isoform expression between some samples. I run Cuffdiff2 and i found the differentially expressed isoforms. I also calculated the porportion of the expression of each isoform compared to the expression of the gene (the sum of the expressions of all the isoforms of that gene). Now I need to better filter my results. I was thinking to focus on the highly expressed isoforms and on those with a relevant proportion, but I couldn't find a criteria to choose a cutoff based on the FPKM. When can I say that a gene or isoform is highly expressed? And how can I understand when a proportion is relevant? I would think that the relevance of the proportion depends also on the total expression, i.e an isoform which account for the 30% of an highly expressed gene might be not as relevant as an isoform which account for the 30% of a low expressed gene, am i correct?
I had a look to related posts in this forum and to several articles but i couldn't find a clear guidline. Maybe the answer is not simple but if you could suggest me some statistical approach to analyze these data would be great.
Thanks in advance!

[Lalla]

FPKM interpretation

Hi all!

I am analyzing differential isoform expression between some samples. I run Cuffdiff2 and i found the differentially expressed isoforms. I also calculated the porportion of the expression of each isoform compared to the expression of the gene (the sum of the expressions of all the isoforms of that gene). Now I need to better filter my results. I was thinking to focus on the highly expressed isoforms and on those with a relevant proportion, but I couldn't find a criteria to choose a cutoff based on the FPKM. When can I say that a gene or isoform is highly expressed? And how can I understand when a proportion is relevant? I would think that the relevance of the proportion depends also on the total expression, i.e an isoform which account for the 30% of an highly expressed gene might be not as relevant as an isoform which account for the 30% of a low expressed gene, am i correct?
I had a look to related posts in this forum and to several articles but i couldn't find a clear guidline. Maybe the answer is not simple but if you could suggest me some statistical approach to analyze these data would be great.
Thanks in advance!

[bastianwur]

There's no good answer for this.
Besides that I'd use the p/q value from the output, that's at least easy .

[Lalla]

Thanks for your reply.
Is not possible at all to do some sort of normalization per abundance in FPKM? If FPKM reflect the expression of transcripts in my sample I think there should be a way to find out the "significance" of a change in relative abundance of a transcript per total expression of a gene. Any suggestion is warmly welcomed!